 So coming up for our third and final of the research partner presentations, we have Curtis Heimerell. He's an assistant professor of computer science at the University of Washington, Go Huskies, working on information and communication technology and international development, specifically universal internet access. Before that, he received his PhD from the University of California at Berkeley, working under professors Eric Brewer and Tappan Perak. I think I pronounced that right. Curtis co-founded Andaga, which joined Facebook in 2015. He was a recipient of the 2014 MIT 35 Under 35 Award and the 2018 UW Early Career Diamond Award. He has won paper awards at CHI, NSDI, Compass, Assets, and Dyspan. Welcome, Curtis. Alright, thanks for having me. I didn't know that would be read out, so I feel bad about it now. Anyway, I know we're always crunched for time, so I will jump into the material. You can see the title has changed slightly. So now I'm going to talk a little bit about opportunities for urban cooperative-style networks, although the content is very similar. Everything aligns very well with some of the other initiatives going on here. It's been really cool to be in the workshop and hear about everything going on, both in our little academic bubbles, as well as in the broader ecosystem. So I'm going to go, I'm going to start a little high-minded. There's been a lot of really technical talks and I wasn't sure exactly where to calibrate this. We get a little bit of both and my job is going to be to sort of tone down the high level and tone up the low level, but I wanted to start with just like the background of the internet a little bit. You know, we like ARPANET existed. It was this big DARPA funded initiative to connect a number of universities, UCLA, UCSD, Utah, UW joined in 1981, so not an early joiner, but up there. And the big idea was really to share resources, right? Supercomputing resources that the NSF had given out and using these new ideas of packet switch networks. The thing that I like to harp on is that it was envisioned as a set of independent autonomous systems peering together at IXPs. Now, that wasn't really the case in 1970, whatever here. But you know, as the network transitioned from a research network to a commercial network, these sort of systems are to get built and you saw things like AS numbers, BGP protocols to support multiple different regional networks working together to connect the country and eventually the world. But if you come to the internet as it is today, it looks like this. These are the top wire line providers in each state. Notably, a midcontinent is actually Comcast. Charter is actually cox if I remember right. And so what you can see is that, you know, 27 million people are on Comcast, 25 million on Charter, and then there's sort of long tail of smaller access. And if you went to, of course, the M&O world, it's even worse. So there's really just three providers now that Sprint has been subsumed. And so, you know, from this, I think, wonderful vision of distributed connectivity, it's sort of boiled down into much simpler systems. And as more content is going over the cellular networks, that particular M&O ecosystem and its lack of competition is becoming problematic. Now, I'm part of a group of researchers, some of the people around, I'm going to put up my chat window to see things. So, I'm part of a group of researchers who have been working on an area called community networks for a long time. These are networks built owned and operated by citizens in a participatory and open manner. Really akin to that model of the old style internet, right. They use primarily inexpensive equipment, unlicensed spectrum, all of this, but their goal at the end of the day is to provide to retake this distributed nature of the internet to provide better service specifically to people on the fringes of current access. People who don't have access, and this is often rural areas, this is often low income urban areas, these kind of places where whatever the existing business processes are, aren't really capable of supporting them all that well. So we've built a couple of these. This is the VBTS Kokomonets project. It's a set of GSM networks deployed in the Philippines with Globe Telecom and the University of the Philippines servicing thousands of people in the remote Aurora province. We've also deployed more modern technologies. This is our current standing LTE installation in Indonesia using a software stack we wrote called Community LTE or Colti, which was built off of OAI originally. And we've deployed a few of these in some other places, but they're continuing to sit there provide IP level connectivity and wide area connectivity to small communities. So the sort of drive of this project was that my student Esther Zhang came to the Philippines to work on the VBTS Kokomonets project but she was originally from New York and she had been volunteering with NYC Mesh, which is one of the largest US cooperative networks in existence. They service Manhattan and Brooklyn through long distance Wi-Fi sort of normal connections ever she had been working with them, went to the Philippines and came back and said hey I think a lot of these cellular models you've been using overseas would actually be applicable to the kind of urban environments that we live in. And so this led us to this question like do we actually want to build these kind of networks in cities. Now you might say okay well no we're living in a city there's a as ubiquitous coverage as there is anywhere on earth here. But if you go a little bit deeper that there might be some lanes here so the city of Seattle actually did and this is of course where we're located. The city of Seattle did a survey here and found that 95% of Seattleites have access to the you know like active connections to the internet. And this seems of course extremely high but as a research group and a researcher who works primarily on technology and poverty alleviation. That last 5% is the problem this is the kind of thing that we work on this is the kind of thing that we beat our head on all the time. And so luckily the city of Seattle and a little bit deeper. And it turns out that it's exactly that so if you are a person who lives in poverty in the city of Seattle. One quarter of those live without internet access. If you're someone who lives with a disability or in a household with a disability 15% don't have internet access so similarly non English speaking older people racial ethnic minorities. All are disproportionately unlikely to be on the internet related to the rest of Seattle so it feels like there may be an opportunity here to target some of these communities with non traditional access idioms and technologies. So why haven't people built these kind of networks before. This is going to be stuff that I think is old hat to a lot of people in the room, but there are really big advances going on that are enabling these kind of smaller networks and smaller access idioms. So, first of course is spectrum. There's things like the citizens broadband radio service this shared license database based back system that's been that went live in in April. And then there's things like LTE you and LTE AA, which are enabling solar to actually run the Wi Fi bands and so chemo has an installation here and our university district doing one of these things. Anyway this used to be a big problem and it's slowing down. There's been a bunch of discussions on Iran and how important that is. Moreover just there's a general commoditization of Iran. Right now you know we were running those two G networks it was $6500 for two G access point 30 maybe even 40 year old technology at this point, but I can buy of course a buy cells, you know be here for $3500. And this is going to continue to go down as we get smaller and smaller cells and things like radio interconnect was always a problem but now that OTT apps are so important we saw this a little bit in Sylvia's talk earlier. A lot of things have been pushed to the edge to make this easier doesn't cover a couple really important use cases, things like emergencies. But because these are IP networks we can hope that the client handles most of this stuff and provides the level of services that the client actually wants for a network to work. And lastly of course, the actual operations element of running these networks now. There are wisps in the real world right now running cellular equipment in the wide area. There are big industry movements towards private LTE we've seen a bunch of that here in the workshop, allowing smaller organizations to run their own LTE networks. And then there's things that carry your aggregation where I as a business or building owner install a whole bunch of access points inside of a building, and lease that out to one of the mno's. And of course magma I think has a place to play in this operations that making it easier for those kind of organizations to, to interact. And so the end story here is to say there's actually a great opportunity right now to take the kind of technologies that we're all talking about and leverage them for these small stakeholder marginalized population connectivity efforts, but there's still a little bit of stuff to figure out. So, that's all kind of background like I said I was going to go real fast through that because I feel like we've gone through a lot of it already. And we'll talk about the solution that we've been building. And the space that it lives in. So the thing that we have is called cooperative cellular. So the codename internally is known as fellow. This is a community access network deployed in the Puget Sound region that people don't know their US geography. So basically everything in the Pacific Northwest that, you know, that's not super close to Portland. So Tacoma, Bainbridge all sort of just our local area but not just Seattle. And the idea here is to do exactly this part to work with local organizations get networks up and build out a telecom here. We see three core components that we've been working on in the lab to, to make this sort of vision this network work. The first one is building out a distributed and federated core network. The second is the spectrum piece for specifically using CBRS at the moment. And the third piece is our community partnerships. I'm going to go deep in the first piece because I think that's got the most technical meat but I'm happy to talk about any of these as they come up for people. Now the idea here is that if we really expect a number of small organizations to run networks, we need to find a way to stitch them together to create the impression or sort of reality of a big wide area network. This is because people, we already have Wi-Fi networks all around everyone's houses what they want is the ability to walk to the local store go to the grocery store maintain their connectivity throughout that. And so this is both authentication and roaming so that they can move between different networks as well as potentially even hand over between those networks. So the vision we have is that you as a user stand up and access point on your roof, you run some software that connects that into the Federation, and then you are able to provide service to your neighbors. Similarly, as you leave the neighborhood, you can join other people's access points and we can build out a wide area city scale and potentially nation scale network through this sort of stitch software. The way that we solve this is through the use of distributed storage that allows us to share critical network state across all of the access points or actually core networks in the Federation, specifically authentication and over and billing information. So little diagram on the bottom kind of shows how we expect the world to work a lot of different networks all peering together to provide this one view of connectivity. So, there's a little bit of background so I want to make sure everyone is in the right headspace. In a traditional cell phone network as you do roaming, your phone will connect to the local MME, which will then through presumably a VPN or some internet connection or maybe even something proprietary, talk to what is your home HS test so if you're somewhere on someone network, you talk to the other person's network, they talk to your home network, provide the authentication credentials and move that over and now this is fine. And it works well in a world where there are very few MNOs because these connections don't don't need to be too many of these connections and of course, almost all marketplaces are dominated by two to five carriers. And so this is how roaming is done, but it won't work for us if we're expecting the thousands of individual carriers to collaborate together. So the way we resolve this problem is to abuse a number of pieces of the LTE authentication setup. And this follows through to 5G as well if people are concerned. So the first thing to note is that we can print our own SIM cards. This is actually really easy if anyone wants to just talk to me about that we buy SIM cards all the time. And so these SIM cards contain the symmetric key of the user. Right. And so they use this in a traditional authentication so that the network sends you a challenge you use the key on your machine to respond to the challenge. And this authenticates the network and the user in both directions. So it turns out as part of this what happens is the your HSS actually creates a thing called an authentication vector. And this is for speeding up authentication. This gets stored on the E node B so that you can move between networks and re authenticate rapidly. Now what we do is we abuse this particular thing by pre generating these authentication vectors before talking to the handset and storing them on this distributed ledger. So in our model user comes and talks to what is the HSS their home network that they got the SIM card from and say hey I need two gigabytes of data. They produce this authentication vector they put it on the on the ledger and that gets shared to all of the PCs throughout the network. Then a user when they show up to another network is able to validate that particular authentication vector and the other HSS is able to do so as well. And so at that point you're authenticated the network you're transiting traffic and you have a certain bill that gets spent down until that reaches zero your two gigabytes are up. At that point you can use either another authentication vector because the user can generate multiple of them, or find another way to get access. What this means is that this is a solution to the way to do a distributed authentication system that's backwards compatible with existing handsets in the book. So we built this. It's working. It's pretty good. Here's our initial results. These are kind of simplistic results. It's basically just saying as we add more nodes we get more capacity but we are able to have users move between networks using this authentication vector mechanism. So okay so that's off that's roaming. Now you should be able to move around different networks inside or different networks inside the Federation. We're similarly working on the problem. Now this is a little bit trickier and I'm not going to go into super deep detail on it because there's a lot here and we're still working on it. But the idea here is the first share similarly neighbor table information on the ledger. So this as well as associated state like location and IP addresses. What this does is it tells all of the other access points and MMEs about everyone inside of the network and all of the access points and allowing them to receive a global vision into the Federated Network. So if the states shared they can then identify each other and put each other on their neighbor tables so that they know that handover the handsets know that handover is possible. We're now then modifying the protocol for adding a protocol for the MMEs to talk to each other to enable handover inside of that. Interestingly, we're also extending this to potentially put the measurement reports from the handsets on the ledger as well. So this is critical if we're going to attack things like unlicensed spectrum because this will allow us to discover Wi-Fi access points that are floating around and choose our spectrum coordination, our channel selection, such that we don't interfere with them even. So a lot of hand waving there, but we're doing handover as well so you can move between the networks. This is the core network piece. Again, we're using CBRS at the moment. I want to just state how surprised they are with how efficient this has been. If people don't know CBRS, it's a big licensed database backend. The idea is that you ask the database for some frequency and it returns them to you and you use the allotment that's been given. We're using Google. They've been gracious and given us access to the system for testing. And so we've just been able to set up gear from there. For example, is the map of available spectrum in the Puget Sound. Down in the bottom, you can see our Tacoma installation. This is one thing that I'll talk about when we get there. But you can see basically there is a lot of spectrum available. Maybe this is because it's in the early days, but it has not been an issue trying to run into other people despite being in a dense urban situation. The biggest problem we have is that there are a number of naval stations near us. And in those cases they may not have given up their frequencies. And so there are parts of the area that are actually just not usable. Lastly, I think, and most importantly, is that you have to work with partners for this kind of work. Deploying the technology, even developing the technology is insufficient when you can't get it into the hands of people that need it, and you can't understand their use cases. So we've been deploying with a couple of local organizations that are really focused on equitable access. These are the Tacoma Cooperative Network, which is a small cooperative network down in Tacoma. Tacoma Public Library is King County Equity Now, which is a social justice organization throughout King County and the city of Seattle's internet for all initiative. I want to specifically call out Althea Networks, which is doing some of the building back end for our system using similar distributed ledger technology. They're really neat. You all should have a look at them as well. So here's our installation on the roof of the gate center on campus. Just as a brief measurement, here's a student of mine, Matt Johnson, you can see this the station pretty far away here. We're able to get 60 megabits per second down and eight megabits per second up at 2.1 kilometers in a dense urban environment. So really the kind of we're able to do much, much better than Wi-Fi installation and provide connectivity throughout a region if we can get the high points. This is our installation in Tacoma. This is on a roof in the hilltop neighborhood that we put up providing connectivity to a small number of nearby families. So we do want to get it out there. We do partner with real organizations and we've some of this magma funding is going exactly to deploying more installations out there. So again, that's just that little green dot down at the bottom. So I'm just about done. You might ask, okay, great. This looks neat. But what does this have to do with magma? And indeed, we worked in the early days with magma specifically about some of the early networking technologies. I think at the end of the day we have similar visions. Our system is, I think, significantly more decentralized in its etiology than magma, but similarly we're trying to provide universal access. So there's a lot of technical overlap. We actually started with the magma fork of OAI, specifically for Colty in Indonesia before its instability just became too problematic for us. We've had great success with the OpenNet 5GS stack. And we're hoping that maybe we can integrate that into the future. But we've started exploring magma based extensions because there are so many sort of functional affordances given by magma that would be really useful for us. We're looking at how to have the orchestrators peer instead of EPCs and whether that would sort of enable a more distributed decentralized version of magma and adding ledger support in general to the orchestrator to peer and collaborate with our existing technical infrastructure. And all of this is still in progress. And there's a lot of sort of threads going on concurrently. But I'm happy to chat more. That's my.